Brian Cheski, co-founder and CEO of Airbnb, defined Design Thinking best as “…starting with the perfect experience and working backwards”. Design Thinking is a powerful approach to designing anything, from services to physical furniture.

So while the actual steps for developing an online accommodation marketplace like Airbnb might differ from designing the iconic Herman Miller’s Aeron chair, Design Thinking offer the same holistic approach to tackling both problems. 

I found Design Thinking principles highly relevant and helpful for understanding and developing custom AI solutions to problems. 

Before diving into implementation details, let’s crystallize three elements:

1. Focus on the end-user. Suppose a physician will use our ML-based application during a surgical procedure. In that case, her practical needs take priority over our previous assumptions and inform critical choices such as the amount of information provided on the fluoroscopy image or the ideal guidance the AI tool offers.
2. Look at the same problem through different “eyes”. People from different backgrounds – anthropologists, engineers, surgeons and designers- bring a “specialized common sense”- an intuitive sound judgment that consistently improves the user experience. 
3. Produce prototypes to iterate on design choices and functionality quickly. Sometimes end-users suggest an idea that adds enormous value to the application or maybe simplifies the problem with their expert knowledge. Often you don’t know what you don’t know.

Implementing the Design Thinking approach

People’s first reaction toward AI often ranges from fearful scepticism to unrealistic expectations driven by current hype. However, even if the above principles help drive the initiative’s success, we must still gain the trust and guide the process ensuring alignment and communication. 

The success of an AI application critically depends on practical, specific circumstances, such as having a clear problem statement, enough data to train models and a flawless workflow to steer the initiative. 

With the experience of multiple projects in different industries, we can suggest a workflow that works well for us.

The workflow

1. Empathize: Begin by understanding the needs and challenges of your clients, as well as the end users of the ML application. Conduct interviews, surveys, and observations to gather insights and develop a deep understanding of their requirements.
2. Define: Based on the gathered insights, define the problem you aim to solve with the ML application. Clearly articulate the goals and objectives of the project, considering both the business and user perspectives. Create a problem statement that focuses on the core issue.
3. Ideate: Brainstorm potential ML solutions that align with the defined problem statement. Encourage creativity and explore various options. Engage in collaborative sessions with stakeholders, including clients, developers, and end users, to generate multiple ideas and concepts.
4. Prototype: Build quick and iterative prototypes of the ML application. These prototypes should represent the key features and functionalities of the proposed solution. The prototypes can be low-fidelity, such as wireframes, mockups, or functional prototypes using simplified ML models.
5. Test: Collect feedback on the prototypes by involving stakeholders and end users. Conduct usability tests and evaluate how well the ML application addresses the defined problem. Use the feedback to refine and iterate on the prototypes, making necessary adjustments and improvements.
6. Implement: Based on the validated prototypes, develop the ML application using appropriate ML algorithms and technologies. Ensure the solution aligns with the requirements identified during the Design Thinking process.
7. Evaluate: Assess the performance and effectiveness of the implemented ML application against the defined objectives. Measure critical metrics, such as accuracy, efficiency, user satisfaction, and business impact. Identify areas for further improvement and gather feedback from users and stakeholders.

Python libraries to use

We strongly believe in removing as much friction and fatigue from developing AI applications, stressing the need to simplify the activities and interactions that support the journey.

Let’s see which well-known libraries can help us:

1. Qualtrics or SurveyMonkey API: the empathize step involves conducting online surveys for large groups. Survey platforms like Qualtrics and SurveyMonkey provide API to integrate survey data collection directly into your Python scripts. These APIs enable you to programmatically create, distribute, and retrieve survey responses.
2. Jupyter Notebook or JupyterLab: Jupyter Notebooks provide an interactive environment for running Python code, documenting your analysis, and sharing your findings. They help present the results during the Define and Evaluate steps and quickly test ideas while brainstorming potential ML solutions.
3. Pandas, Matplotlib and Seaborn: these libraries enable you to organize, clean, analyze data and create visualizations and plots. I can’t stress enough how visualizations can aid in conveying information effectively and identifying patterns or trends in the data.
4. PyTorch Lightning: Lightning is the PyTorch research framework that I use for the Prototype and Test phases. I love Lightning as it makes me focus on research and test ideas, accelerating system development. Furthermore, Lightning helps me build systems that are scalable and ready for production.  
5. PySerial: PySerial is a library for serial communication in Python. It lets you interact with serial devices or interfaces, such as Arduino boards or sensors, from your Python code. PySerial can be helpful when developing prototypes that interact with hardware components.

Ensuring a smoother development experience

AI is today an impressively fast-paced environment that requires a culture of continuous learning and innovation to embrace emerging technologies and industry best practices. Therefore, we must continuously invest in research and development to explore new AI techniques, frameworks, and applications. 

However, it should be clear by now that developing Artificial Intelligence systems is not (only) software development and data cleaning. From gaining a thorough understanding of the client’s business goals to fostering a collaborative relationship with clients, many “softer” aspects become critical success factors. 

In this challenging, exciting industry, sound Design Thinking principles and the above workflow helped me prevent common issues with AI developments and ensure a smoother development experience.